March 2026 Latest

LTX-2.3 ComfyUI

Unlock the peak of open-source video generation. From basic templates to professional workflows.

Step-by-Step Running Guide

Step 1: Environment

Install "ComfyUI-LTXVideo" via Manager and restart.

Step 2: Load Models

Put models in correct paths and refresh.

Step 3: Import Workflow

Drag & Drop JSON into ComfyUI browser.

Model Installation Paths

Checkpoints (UNet)
ComfyUI/models/checkpoints/LTX-Video/
Upscale Models
ComfyUI/models/upscale_models/
LoRAs
ComfyUI/models/loras/

Why Run LTX-2.3 in ComfyUI?

ComfyUI is the most powerful platform for LTX-2.3. Control two-stage upscaling, audio sync, and multi-modal ControlNet with ease.

Deep Dive: Two-Stage Workflow

The recommended way for Distilled models to balance speed and high-frequency details.

Stage 1: Latent Base

Generate global motion and structure at lower resolution (e.g., 544x960).

Stage 2: Spatial Upscale

Inject high-frequency textures and sharpen edges using spatial-upscaler.

Hardware Requirements

Entry-Level (8-12GB VRAM)

Use GGUF/FP8. Recommended: 720p vertical, single stage.

Mainstream (16-24GB VRAM)

Best for Two-Stage. 1080p output in 30-60s.

Production (24GB+ VRAM)

Run bf16 Dev full model. Supports 4K spatial + temporal upscale.

Prompt Library (Copyable)

Cinematic Vertical

"A low-angle close-up shot of a cyberpunk detective walking through a neon-lit rain-slicked street, volumetric lighting, 8k resolution, cinematic grain, sharp textures."

Nature I2V

"A majestic waterfall cascading down a lush green mountain, early morning mist, slow-motion drone shot, hyper-realistic, 4k, sunlight filtering through trees."

Dynamic Dance

"A professional dancer performing hip-hop on a rooftop at sunset, fluid motion, dynamic camera movement, vibrant colors, shallow depth of field."

1. Official Built-in Templates

Built-in templates for quick start.

Steps:

  • Update ComfyUI to latest (≥ 0.16.1).
  • Menu -> Template Library, search "LTX-2.3".
  • Load T2V or I2V template.

Troubleshooting (FAQ)

Q: "Model Not Found"?

A: Check if models are in LTX-Video subfolder with .safetensors extension.

Q: Black screen or noisy video?

A: Distilled model requires CFG=1.0. Dev model suggests 3.0-5.0.

Q: Out of Memory?

A: Reduce resolution or use GGUF quantized versions.

Key Parameters (Distilled)

Steps: 8-20
CFG: 1.0 - 3.0 (1.0 recommended for Distilled)
Frames: 8n + 1 (e.g., 25, 49)
Resolution: 832x1472 or 960x544

"LTX-2.3 is the game changer for open-source AI video in ComfyUI."

© 2026 wan2.video